I’m very proud of this little paper called Slope One Predictors for Online Rating-Based Collaborative Filtering. The paper report on some of the core collaborative filtering research leading to the inDiscover web site. I’ll be presenting it at SIAM Data Mining 2005 in April (Newport Beach, California).
This is a case where, with Anna Maclachlan, we did something that few researchers do these days: we looked for something simpler. The main result of the paper is that you can use extremely simple and easy to implement algorithms and get very competitive results.
The current trend, in academia, is to develop crazy algorithms that require not 10 lines of code, not 100 lines of code, but several thousands. I think the same is true in some industries: think of Web Services or Java (with the infinite number of new acronyms).
Well, I like complex algorithms and as a math guy, I like a challenge, but once in a while, I think it pays to go I think “wait! what if the average Joe wants to implement this?”
So, if you write real code and are interested in collaborative filtering, go check this paper.
You wouldn’t have open code for this lying anywhere, would you?
Valid request. I’ll post something in the near future. I do have lots of code, but it is Java code, I think I should post SQL code.
Yes, some SQL code would be fantastic! Maybe with some pseudo-code for some of the logic… Something us non-math geeks can understand 🙂
oh oh, looks like I’m not getting any sleep soon!
For code samples, go there
http://www.daniel-lemire.com/fr/abstracts/TRD01.html
It’s beauty is its simplicity.
Great paper!